KR102373642B1 - Chromosome analysis system for diagnosis of radiation exposure based on deep learning - Google Patents

Abstract

The present invention relates to a chromosome reading system for diagnosing radiation exposure based on deep learning, and to a chromosome reading system for diagnosing radiation exposure, comprising: a first neural network for recognizing chromosomes; a second neural network for recognizing abnormal and duplicate chromosomes among the chromosomes; an analysis unit including a first analysis unit for analyzing the number of chromosomes and a second analysis unit for analyzing the number of abnormal and duplicate chromosomes among the analyzed chromosomes; and a reading unit that determines that the chromosome is unreadable, normal, or abnormal based on the results of the first analysis unit and the second analysis unit. By using the chromosome reading system for diagnosing radiation exposure, it is possible to read radiation-exposed chromosomes with high accuracy of 80% or more while shortening the reading time by 1/10 compared to the prior art.

Description

Chromosome reading system for diagnosing radiation exposure based on deep learning {CHROMOSOME ANALYSIS SYSTEM FOR DIAGNOSIS OF RADIATION EXPOSURE BASED ON DEEP LEARNING}

The present invention relates to a chromosome reading system for diagnosing radiation exposure based on deep learning. Specifically, in order to improve chromosome reading accuracy, each chromosome and abnormal and duplicate chromosomes among the chromosomes are recognized using the first and second neural networks, and the number of chromosomes is analyzed using two neural networks. Or it relates to a chromosome reading system that is used to diagnose the degree of radiation exposure by judging it as abnormal.

As the analysis of the effects of radiation exposure on the human body begins, methods for the evaluation of radiation exposure are also gradually developed. In particular, it is necessary to quantitatively evaluate the radiation exposure of the human body in the case of a radiation accident that is unintentionally exposed to radiation and radioactive materials. is the majority

In the case of a radiation exposure accident, the structural destruction of DNA occurs by radiation, thereby causing abnormal chromosomal abnormalities such as moving source chromosomes. Biological dose evaluation methods for estimating exposure dose by analyzing abnormal chromosomes (cells) include cytogenetic method (international standard method), moving source chromosome analysis method, translocation analysis method, micronucleus analysis method, and immature chromosome condensation analysis method.

In order to use the moving source chromosome analysis method, which is mainly used in acute exposure accidents, for the evaluation of biological dose, blood collection, lymphocyte isolation and culture, cell recovery, cell fixation, slide production, slide staining, metaphase cell selection, and chromosome reading are performed. In the chromosome reading process, it is common for a person to directly read the presence or absence of chromosomal abnormalities with respect to individual samples, and the number of metaphase cells that can be read by one person per day is about 500. In particular, it requires a lot of time and manpower to read chromosomes because it is necessary to evaluate the degree of exposure by counting 1,000 metaphase cells for each test sample. In addition, when a large nuclear power plant accident such as the Chernobyl accident or the Fukushima accident occurs, it is very difficult to utilize it in terms of time and cost if it is necessary to perform a biological dose assessment on a group population.

To date, the only commercial program developed for automated chromosome reading is DCScore software provided by MetaferSystems. is coming However, this program has a disadvantage in that the accuracy is somewhat lower compared to the analysis standards (counting chromosome number and division of moving source chromosomes) suggested by the naked eye. In particular, the conventional chromosome reading system for radiation exposure uses a computer vision (CV) method for chromosome recognition, etc., and has a problem of low accuracy by staying at the level of introducing a machine learning method for some readings.

U.S. Patent Publication No. US8605981 (2013.12.10) describes a patent for a method for detecting abnormal chromosomes and evaluating radiation dose. In this patent, chromosomes are separated using a threshold value for each pixel of each image (Segmentation). Centromere is detected in each separated chromosome, and whether centromere (abnormal chromosome having two or more centromere) is detected using individually detected centromere SVM or machine learning method of artificial neural network. In addition, the process of estimating the exposure dose based on the detected abnormal (moving source) chromosome number coefficient, etc. is included. The method was developed as a program to which artificial neural network technology was applied in some courses and is currently being used by various institutions. However, when applied in practice, the problem of low accuracy in the counting of chromosome numbers and recognition of abnormal (moving source) chromosomes arises, and in most cases, a human must go through the process of re-verification. In addition, there is an error that may indicate a large difference from the naked eye because the method of checking only the abnormal number of chromosomes is taken in the standard that does not count the number of chromosomes.

In "Automated Discrimination of Dicentric and Monocentric Chromosomes by Machine Learning-based Image Processing" published in 2016, we propose a method for recognizing normal chromosomes and migratory chromosomes using the SVM machine learning method. The SVM method received attention as a machine learning method with considerably high accuracy before the development of deep learning technology, but there are limitations in the accuracy of the SVM method and the accuracy of chromosome reading as it uses the segmentation method to recognize individual chromosomes.

In the paper "Accurate cytogenetic biodosimetry through automated dicentric chromosome curation and metaphase cell selection" published in 2017, after separating chromosomes using image segmentation technology, the centromere The source of the movement) reads whether there is a chromosome or not. It is judged that the accuracy is improved compared to the existing method, but it is a method that developed the image processing technique (Computer Vision, CV) before the deep learning technology, and it is difficult to derive the maximum efficiency in the image reading process where there are many variables.

"A dicentricchromosome identification method based on clustering and watershed algorithm" published in 2019 relates to a method of counting normal chromosomes and migratory chromosomes using an image processing method and performing dose evaluation. It showed a significant improvement in chromosome processing technology including nested chromosomes, and achieved a reading accuracy approaching 80%, but this also showed limitations in image reading accuracy using an image processing technique (CV).

Therefore, in order to quickly perform a radiation exposure diagnosis in preparation for a radiation accident occurring in a group, there is an urgent need for a method with high accuracy of chromosome reading and shortening the reading time.

The present application relates to a chromosome reading system for diagnosing radiation exposure to solve the problems of the prior art, and it is possible to increase reading accuracy and reduce reading time at the same time through biological dose evaluation using deep learning-based object recognition.

Specifically, in order to improve chromosome reading accuracy, each chromosome, abnormal and duplicate chromosomes among the chromosomes are recognized using the first and second neural networks, and the number of chromosomes is analyzed using two neural networks, and the chromosomes are unreadable, normal , or to a radiation-exposed chromosome reading system that is judged to be abnormal.

A chromosome reading system for diagnosing radiation exposure based on deep learning of the present invention for achieving the above technical problem includes: a first neural network for recognizing chromosomes; a second neural network for recognizing abnormal and duplicate chromosomes among the chromosomes; an analysis unit including a first analysis unit for analyzing the number of chromosomes and a second analysis unit for analyzing the number of abnormal and duplicate chromosomes among the analyzed chromosomes; and a reading unit that determines that the chromosome is unreadable, normal, or abnormal based on the results of the first analysis unit and the second analysis unit.

The first neural network, a first labeling unit for labeling the chromosome; a first data enlargement unit for enlarging the first data of the chromosome; and a first learning unit that educates the first data through machine learning, but is not limited thereto.

The second neural network may include: a second labeling unit for labeling the abnormal and the duplicate chromosomes; a second data enlargement unit for enlarging second data of the abnormal and duplicate chromosomes; and a second learning unit that educates the second data through machine learning, but is not limited thereto.

The first analysis unit may analyze the number of chromosomes using the results, thresholds, and Equation 1 below of the first and second neural networks, but is not limited thereto.

[Equation 1]

In Equation 1, N _T is the number of chromosomes, N ₁ is the number of chromosomes recognized through the first neural network, and N _i (S ₁₂ |C) is the region recognized through the second neural network The region S ₁₂ recognized by the first neural network and the i number of duplicated chromosomes satisfying the criterion C may be the number of chromosome objects included in a single object, but is not limited thereto.

The second analysis unit may analyze the number of abnormal and duplicate chromosomes using the results of the first and second neural networks and Equation 2 below, but is not limited thereto.

[Equation 2]

In Equation 2, D _F (Ab) is the number of abnormal chromosomes, D ₁ (Cr) is the number of chromosomes recognized through the first neural network, and D ₂ (Ab) is the number of chromosomes recognized through the second neural network. It may be the number of recognized abnormal chromosomes, but is not limited thereto.

The reading unit cannot read if the number of chromosomes does not satisfy Equation 3 below, and if the number of chromosomes satisfies Equation 3 below and includes an abnormal chromosome, it is abnormal, and if it does not, it is determined to be normal. , but is not limited thereto.

[Equation 3]

In Equation 3, N may be the number of chromosomes, and I may be 0 to 5, but is not limited thereto.

A chromosome reading method for diagnosing radiation exposure includes: recognizing, by a chromosome reading system, a chromosome; recognizing abnormal and redundant chromosomes; analyzing the number of normal, abnormal and duplicate chromosomes based on the recognized information; and determining the chromosome as unreadable, normal or abnormal using the analyzed information.

Recognizing the chromosome may include: labeling the chromosome; enlarging the data of the chromosome; and educating the first data through machine learning; may include, but is not limited thereto.

Recognizing the abnormal and redundant chromosomes may include: labeling the abnormal and redundant chromosomes; enlarging the data of the abnormal and the duplicated chromosomes; and educating the second data through machine learning, but is not limited thereto.

The step of analyzing the number of chromosomes may be using Equation 1 below, but is not limited thereto.

[Equation 1]

In Equation 1, N _T is the number of chromosomes, N ₁ is the number of recognized chromosomes, and N _i (S ₁₂ | C) is the region recognized in the step of recognizing the abnormal and duplicate chromosomes. The region (S ₁₂ ) recognized in the step of recognizing the chromosome and the i number of duplicated chromosomes satisfying the criterion C may be the number of chromosome objects included in a single object, but is not limited thereto.

The step of analyzing the number of abnormal and duplicate chromosomes may be performed using Equation 2 below, but is not limited thereto.

[Equation 2]

In Equation 2, D _F (Ab) is the number of abnormal chromosomes, D ₁ (Cr) is the number of chromosomes recognized in the step of recognizing the chromosome, and D ₂ (Ab) is the abnormal and duplicate chromosomes It may be the number of abnormal chromosomes recognized in the step of recognizing, but is not limited thereto.

In the determining step, if the number of chromosomes does not satisfy Equation 3 below, it is impossible to read, and if the number of chromosomes satisfies Equation 3 below and contains an abnormal chromosome, it is abnormal, and if it does not, it is determined as normal. However, the present invention is not limited thereto.

[Equation 3]

In Equation 3, N may be the number of chromosomes, and I may be 0 to 5, but is not limited thereto.

The above-described problem solving means are merely exemplary, and should not be construed as limiting the present application. In addition to the exemplary embodiments described above, additional embodiments may exist in the drawings and detailed description.

The disclosed technology may have the following effects. However, this does not mean that a specific embodiment should include all of the following effects or only the following effects, so the scope of the disclosed technology should not be understood as being limited thereby.

According to the above-mentioned means for solving the problems of the present application, the deep learning-based radiation exposure diagnosis chromosome reading system according to the present application uses machine learning and two neural networks to improve the accuracy (80% or more) than the conventional chromosome determination system. , it is possible to reduce the reading time by more than 1/10 compared to the conventional method by performing automatic reading using machine learning.

In addition, although one person has to analyze one sample in visual reading, when the reading system according to the present invention is used, one person can analyze several samples at the same time, so the inspection efficiency can be increased.

1 is a diagram of a chromosome reading system for diagnosing radiation exposure based on deep learning according to an embodiment of the present application.
2 is a diagram of an algorithm for determining a threshold according to an embodiment of the present application.
3 is a diagram of a chromosome reading system algorithm for diagnosing radiation exposure based on deep learning according to an embodiment of the present application.
4 is a flowchart of a chromosome reading method for diagnosing radiation exposure based on deep learning according to an embodiment of the present application.
5 is a photograph in which the number of chromosomes measured according to the present embodiment is counted.
6 is a photograph in which the number of abnormal and duplicate chromosomes measured according to the present embodiment is counted.

Since the present invention can have various changes and can have various embodiments, specific embodiments are illustrated in the drawings and will be described in detail in the detailed description. However, this is not intended to limit the present invention to specific embodiments, and it should be understood to include all modifications, equivalents, and substitutes included in the spirit and scope of the present invention.

In describing each figure, like reference numerals are used for like elements. Terms such as first, second, etc. may be used to describe various elements, but the elements should not be limited by the terms. The above terms are used only for the purpose of distinguishing one component from another.

For example, without departing from the scope of the present invention, a first component may be referred to as a second component, and similarly, a second component may also be referred to as a first component. The term “and/or” includes a combination of a plurality of related listed items or any of a plurality of related listed items.

Unless defined otherwise, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

Terms such as those defined in a commonly used dictionary should be interpreted as having a meaning consistent with the meaning in the context of the related art, and should not be interpreted in an ideal or excessively formal meaning unless explicitly defined in the present application. shouldn't

Throughout this specification, when a member is positioned “on”, “on”, “on”, “on”, “under”, “under”, or “under” another member, this means that a member is positioned on the other member. It includes not only the case where they are in contact, but also the case where another member exists between two members.

Throughout this specification, when a part "includes" a certain component, it means that other components may be further included, rather than excluding other components, unless otherwise stated.

As used herein, the terms "about," "substantially," and the like are used in a sense at or close to the numerical value when the manufacturing and material tolerances inherent in the stated meaning are presented, and to aid in the understanding of the present application. It is used to prevent an unconscionable infringer from using the mentioned disclosure in an unreasonable way. Also, throughout this specification, "step to" or "step to" does not mean "step for".

Throughout this specification, the term "combination of these" included in the expression of the Markush form means one or more mixtures or combinations selected from the group consisting of the components described in the expression of the Markush form, and the components It is meant to include one or more selected from the group consisting of.

Hereinafter, the radiation-exposed chromosome reading apparatus of the present application and the radiation-exposed chromosome reading method using the same will be described in detail with reference to embodiments, examples, and drawings. However, the present application is not limited to these embodiments and examples and drawings.

The present application, a first neural network for recognizing a chromosome; a second neural network for recognizing abnormal and duplicate chromosomes among the chromosomes; an analysis unit including a first analysis unit for analyzing the number of chromosomes and a second analysis unit for analyzing the number of abnormal and duplicate chromosomes among the analyzed chromosomes; It relates to a chromosome reading system for confirming a diagnosis of radiation exposure, including a reading unit that determines that a chromosome is unreadable, normal, or abnormal based on the results of the first analysis unit and the second analysis unit.

The radiation-exposed chromosome reading apparatus according to the present disclosure may be used in the chromosome reading process when biological dose evaluation is performed by the moving-source chromosome analysis method. In addition, by utilizing the technology of the present application, it can be utilized in other fields of chromosomal reading.

1 is a diagram of a chromosome reading system for diagnosis of deep learning-based radiation exposure according to an embodiment of the present application.

Specifically, the radiation exposure chromosome reading system according to the present application (100) is a first neural network 110 for recognizing chromosomes, a second neural network 120 for recognizing abnormal and duplicate chromosomes among the chromosomes, and a first for analyzing the number of chromosomes. An analysis unit 130 including a first analysis unit 131 and a second analysis unit 132 for analyzing the number of abnormal and duplicate chromosomes among the analyzed chromosomes, the first analysis unit 131 and the second analysis unit A reading unit 140 that determines the chromosome as unreadable, normal, or abnormal as the result of the unit 132 is included.

The first neural network 110, a first labeling unit (not shown) for labeling the chromosome; a first data enlargement unit (not shown) for enlarging the first data of the chromosome; and a first learning unit (not shown) for educating the first data through machine learning, but is not limited thereto.

The chromosome recognized by the first neural network 110 includes all chromosomes regardless of normal, abnormal, and duplicate chromosomes. The first labeling unit of the first neural network 110 labels each recognized chromosome. Next, the data is enlarged in the first data enlarger in order to increase learning efficiency. In this case, the first data enlargement unit may perform rotation, flip, resizing, and combinations thereof on the data. Specifically, in order to accurately read the chromosome recognized in the image, the chromosome is rotated, flipped, resizing, etc. to make the chromosome more clearly recognized. Then, the processed first data is trained by machine learning (machine learning and deep learning) in the first learning unit. The reading accuracy of the chromosome may be improved by learning the first data in the first learning unit. In addition, reading accuracy may be improved through machine learning using the learned first data. The first neural network 110 constructed through learning may quantify the number, position, size, etc. of the individually recognized chromosomes.

The first neural network 110 may use a neural network selected from the group consisting of a Feature Pyramid Network, a Faster RcNN algorithm, RetinaNet, and combinations thereof, but is not limited thereto.

The machine learning (Machine Learning) is a type of artificial intelligence, which means that a computer learns by itself through given data. Machine learning includes functions and generalizations for representing data and evaluating them. Generalization means that the current model is applied to new data as it is.

The second neural network 120 includes a second labeling unit (not shown) for labeling the abnormal and the duplicate chromosomes; a second data enlargement unit (not shown) for enlarging second data of the abnormal and duplicate chromosomes; and a second learning unit (not shown) for educating the second data through machine learning, but is not limited thereto.

The abnormal chromosome may include an abnormal chromosome selected from the group consisting of Dicentric Chromosome, Acentric Chromosome, Ring Chromosome, and combinations thereof. it is not

The second neural network 120 recognizes the abnormal and the duplicate chromosome. Although the first neural network 110 can perform multiple object recognition, in the present application, the second neural network 120 recognizes the abnormal and the duplicate chromosomes, thereby improving the reading accuracy of the chromosomes. The second labeling unit of the second neural network 120 labels each recognized chromosome. Next, the data is enlarged in the second data enlarger to increase learning efficiency. In this case, the second data enlargement unit may perform rotation, flip, resizing, and combinations thereof on the data. Specifically, in order to accurately read the chromosome recognized in the image, the chromosome is rotated, flipped, resizing, etc. to make the chromosome more clearly recognized. Next, the processed second data is trained by machine learning (machine learning and deep learning) in the second learning unit. The reading accuracy of the chromosome may be improved by learning the second data in the second learning unit. In addition, reading accuracy may be improved through machine learning using the learned second data. The second neural network 120 constructed through learning may quantify the number, position, size, etc. of the individually recognized chromosomes.

The second neural network 120 may use a neural network selected from the group consisting of a Feature Pyramid Network, a Faster RcNN algorithm, RetinaNet, and combinations thereof, but is not limited thereto.

Chromosomes can be recognized more accurately by recognizing each of the chromosomes using the two neural networks of the first neural network 110 and the second neural network 120 at the same time than when the neural network is used alone.

The first analysis unit 131 may analyze the number of chromosomes using the results of the first neural network 110 and the second neural network 120, a threshold value, and Equation 1 below. It is not limited.

In Equation 1, N _T is the total number of chromosomes, N ₁ is the number of chromosomes recognized through the first neural network, and N _i (S ₁₂ |C) is a region recognized through the second neural network The region S ₁₂ recognized by the first neural network and the i number of duplicated chromosomes satisfying the criterion C may be the number of chromosome objects included in a single object, but is not limited thereto.

Specifically, when the overlapping degree of the area (image) recognized by the neural network 1 and the area recognized by the neural network 2 is expressed in %, when it is greater than the reference C%, the overlapping area of the neural network 1 and the neural network 2 It may be to select i chromosomes in

In Equation 1, the chromosomes recognized through the first neural network 110, that is, duplicated chromosomes by correcting using the values of the chromosomes recognized through the second neural network in all numbers N ₁ of normal, abnormal, and duplicate chromosomes. The total number of chromosomes N _T can be obtained by correcting the number of .

The threshold value may be determined through the algorithm of FIG. 2 .

The threshold is a reference value of the probability of recognizing an object (chromosome) as a correct answer when the first neural network 110 or the second neural network 120 detects an object (chromosome), and may be an IOU Threshold or a Score Threshold. In general, when the threshold is set low, the detection rate may increase, and when the threshold is set high, the detection rate may decrease.

2 is a diagram of an algorithm for determining a threshold according to an embodiment of the present application.

Referring to FIG. 2 , the threshold value is set such that the number N of the chromosomes is N=46±I. Specifically, in FIG. 2 , T is an actually applied threshold value, T _c is an initial value of the threshold value, and t is an amount of change of the threshold value, which may be about 0.01, but is not limited thereto. The T _c may be determined by learning in the first neural network 110 and/or the second neural network 120 .

In addition, I may be 0 to 5 as a tolerance for the number of chromosome reads, but is not limited thereto. When I is 0, N is 46, which is the number of chromosomes of a normal person. However, when the I is greater than 5, the number of chromosomes may have a difference compared to the number of chromosomes of a normal person, so that it may be difficult to make an accurate reading.

The deep learning-based chromosome reading system for diagnosing radiation exposure according to the present application can secure the optimal chromosome recognition probability by automatically setting the threshold value through the first neural network 110 and/or the second neural network 120 .

The second analysis unit 132 may analyze the number of abnormal and duplicate chromosomes using the results of the first neural network 110 and the second neural network 120 and Equation 2 below, However, the present invention is not limited thereto.

In Equation 2, D _F (Ab) is the number of abnormal chromosomes, D ₁ (Cr) is the number of chromosomes recognized through the first neural network, and D ₂ (Ab) is the number of chromosomes recognized through the second neural network. It may be the number of recognized abnormal chromosomes, but is not limited thereto.

There are cases in which the abnormal chromosome recognized by the second neural network 120 is recognized even when it is not a chromosome, and in order to minimize this error, the number of abnormal chromosomes is corrected through Equation 2 above.

The reading unit 140 cannot read if the number of chromosomes does not satisfy Equation 3 below, and if the number of chromosomes satisfies Equation 3 and contains an abnormal chromosome, it is abnormal, and if it does not, it is determined as normal. may be, but is not limited thereto.

In Equation 3, N may be the number of chromosomes, and I may be 0 to 5, but is not limited thereto.

The I is the tolerance of the number of chromosome reads.

The reading unit 140 may read radiation-exposed chromosomes according to the algorithm of FIG. 3 .

3 is a diagram of a radiation exposure chromosome determination algorithm according to an embodiment of the present application.

Referring to FIG. 3 , it is first determined whether Equation 3 is satisfied based on the number of chromosomes analyzed by the analysis unit in a chromosome image for reading chromosomes. Since there are 23 pairs of human chromosomes, a total of 46, it is first determined whether the chromosomes exist within the tolerance range. In the conventional chromosome reading (US Patent Publication No. US8605981B2), since the number of chromosomes is not counted and only the number of abnormal chromosomes is checked, it is different from reading with the naked eye. Specifically, when only 10 chromosomes are detected, if only normal chromosomes are detected, the prior art is determined to be normal, whereas in the present invention, it will be determined to be unreadable. However, there is a total of 46 chromosomes in humans, and when 10 of them are normal, there is an error in judging that the remaining 36 chromosomes are also normal. That is, the radiation-exposed chromosome reading apparatus according to the present application can reduce errors and increase data reliability by first counting chromosomes.

Next, when the counted number of chromosomes satisfies Equation 3, it is determined that the abnormal chromosome is included in the abnormal case and that the abnormal chromosome is not included in the normal case.

The radiation exposure dose can be determined based on the result read by the reading unit 140 , and the radiation dose can be directly estimated through additional processing.

The accuracy of the radiation-exposed chromosome determination device may be 80% to 100%, but is not limited thereto.

In the case of a conventional chromosome determination system or program, the accuracy did not exceed 80%, whereas the radiation exposure chromosome determination apparatus according to the present application exhibits an accuracy of 80% or more. Specifically, the read rate of individual chromosomes is 95% or more, and the read rates of abnormal and duplicate chromosomes are 95% or more. Moreover, it can be expected that the accuracy will be further increased through continuous learning.

The deep learning-based chromosome reading system for diagnosing radiation exposure according to the present application uses machine learning and two neural networks to improve accuracy compared to the conventional chromosome judgment system, and by performing automatic reading by machine learning, it is better than the conventional visual reading method. The reading time can be shortened by 1/10 or more compared to the prior art. In addition, since a large number of samples can be analyzed at the same time, the inspection efficiency can be increased.

Herein, a deep learning-based chromosome reading system for diagnosing radiation exposure comprises: recognizing a chromosome; recognizing abnormal and redundant chromosomes; analyzing the number of normal, abnormal, duplicate and unreadable chromosomes based on the recognized information; and judging the chromosome as unreadable, normal, or abnormal using the analyzed information; provides a chromosome reading method for radiation exposure diagnosis comprising a.

4 is a flowchart of a reading method using a deep learning-based chromosome reading system for diagnosing radiation exposure according to an embodiment of the present application.

First, the chromosome is recognized (S100).

Recognizing the chromosome may include: labeling the chromosome; enlarging the data of the chromosome; and educating the first data through machine learning; may include, but is not limited thereto.

The chromosome includes all chromosomes regardless of normal, abnormal, and duplicate chromosomes.

The step of recognizing the chromosome may be that the first neural network 110 recognizes the chromosome, but is not limited thereto. Specifically, the first labeling unit may label the chromosome, the first data expansion unit may expand the data of the chromosome, and the first learning unit may educate the first data, but is not limited thereto.

Subsequently, abnormal and duplicate chromosomes are recognized (S200).

Recognizing the abnormal and redundant chromosomes may include: labeling the abnormal and redundant chromosomes; enlarging the data of the abnormal and the duplicated chromosomes; and educating the second data through machine learning, but is not limited thereto.

The step of recognizing the abnormal and duplicate chromosomes may be that the second neural network 120 recognizes the abnormal and duplicate chromosomes, but is not limited thereto. Specifically, the second labeling unit labels the abnormal and duplicate chromosomes, the second data expansion unit expands the data of the abnormal and duplicate chromosomes, and the second learning unit may educate the second data, However, the present invention is not limited thereto.

Next, the number of normal, abnormal, and duplicate chromosomes is analyzed based on the recognized information (S300).

The step of analyzing the number of chromosomes may be using Equation 1 below, but is not limited thereto.

[Equation 1]

In Equation 1, N _T is the number of chromosomes, N ₁ is the number of recognized chromosomes, and N _i (S ₁₂ | C) is the region recognized in the step of recognizing the abnormal and duplicate chromosomes. The region (S ₁₂ ) recognized in the step of recognizing the chromosome and the i number of duplicated chromosomes satisfying the criterion C may be the number of chromosome objects included in a single object, but is not limited thereto.

The analyzing of the number of chromosomes may be performed by using the results of the first neural network 110 and the second neural network 120, the threshold value, and Equation 1, but is not limited thereto.

The step of analyzing the number of abnormal and duplicate chromosomes may be performed using Equation 2 below, but is not limited thereto.

[Equation 2]

In Equation 2, D _F (Ab) is the number of abnormal chromosomes, D ₁ (Cr) is the number of chromosomes recognized in the step of recognizing the chromosome, and D ₂ (Ab) is the abnormal and duplicate chromosomes It may be the number of abnormal chromosomes recognized in the step of recognizing, but is not limited thereto.

The analyzing of the number of abnormal and duplicate chromosomes may be performed by using the results of the first neural network 110 and the second neural network 120 and Equation 2, but is not limited thereto.

Then, it is determined that the chromosome is unreadable, normal, or abnormal using the analyzed information (S400).

In the determining step, if the number of chromosomes does not satisfy Equation 3 below, it is impossible to read, and if the number of chromosomes satisfies Equation 3 below and contains an abnormal chromosome, it is abnormal, and if it does not, it is determined as normal. However, the present invention is not limited thereto.

[Equation 3]

In Equation 3, N may be the number of chromosomes, and I may be 0 to 5, but is not limited thereto.

In the determining step, by first performing chromosome counting, errors can be reduced and data reliability can be increased.

The present invention will be described in more detail through the following examples, but the following examples are for illustrative purposes only and are not intended to limit the scope of the present application.

First, 10 ml of venous blood, which is peripheral blood, was collected from the subject using aseptic treatment and a heparin-treated syringe. Lymphocytes were separated by centrifugation by placing the collected venous blood on the Ficoll-Hypaq solution injected into the test tube. The LCM 10 culture solution was added to the isolated lymphocytes, washed twice, and the LCM 10 culture solution was added again so that the number of cells became 4x10 ⁶ . After centrifugation again, 6 ml of 0.075 M KCl 5.6 g/l stock solution was added, left for 10 minutes, and centrifuged to remove the supernatant. Then, after fixing three times with Carnoy's fixative (methanol:acetic acid=3:1), the fixed cells were dropped onto slides, dried in air, and then 5% Giemsa staining was performed. After obtaining an image of the dyed chromosome, it was read by the deep learning-based radiation exposure diagnosis chromosome reading system according to the present application.

5 is a photograph in which the number of chromosomes measured according to the present embodiment is counted.

5 is a photograph showing the number of chromosomes counted by the first analysis unit of the deep learning-based radiation exposure diagnosis chromosome reading system. As a result of reading 130 images, the reading accuracy was more than 95% based on individual chromosomes.

6 is a photograph in which the number of abnormal and duplicate chromosomes measured according to the present embodiment is counted.

6 is a photograph showing the number of abnormal and duplicate chromosomes counted by the second analysis unit of the radiation-exposed chromosome reading apparatus. As a result of reading 196 images, it was found that the reading accuracy of each of the duplicate and abnormal chromosomes was 85% or more.

According to the results shown in the above examples, it was confirmed that the radiation-exposed chromosome reading apparatus of the present application exhibits an accuracy of 80% or more, which is superior to that of the existing ones in reading chromosomes.

The above description of the present application is for illustration, and those of ordinary skill in the art to which the present application pertains will understand that it can be easily modified into other specific forms without changing the technical spirit or essential features of the present application. Therefore, it should be understood that the embodiments described above are illustrative in all respects and not restrictive. For example, each component described as a single type may be implemented in a dispersed form, and likewise components described as distributed may also be implemented in a combined form.

The scope of the present application is indicated by the following claims rather than the above detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalent concepts should be construed as being included in the scope of the present application.

100: Deep Learning-based Chromosome Reading System for Diagnosis of Radiation Exposure
110: first neural network
120: second neural network
130: analysis unit
131: first analysis unit
132: second analysis unit
140: reading unit

Claims (12)

a first neural network for recognizing chromosomes;
a second neural network for recognizing abnormal and duplicate chromosomes among the chromosomes;
an analysis unit including a first analysis unit for analyzing the number of chromosomes and a second analysis unit for analyzing the number of abnormal and duplicate chromosomes among the analyzed chromosomes;
and a reading unit that determines that the chromosome is unreadable, normal, or abnormal based on the results of the first analysis unit and the second analysis unit;
The first analysis unit is to analyze the number of chromosomes using the results of the first neural network and the second neural network, a threshold value, and the following Equation 1,
The second analysis unit analyzes the number of the abnormal and the duplicate chromosomes using the results of the first and second neural networks and the following Equation 2, a deep learning-based radiation exposure diagnosis chromosome reading system:
[Equation 1]

(In Equation 1 above,
The N _T is the number of chromosomes,
The N ₁ is the number of chromosomes recognized through the first neural network,
The N _i (S ₁₂ |C) is a chromosome in which the region recognized through the second neural network meets the criterion C and the region recognized by the first neural network (S ₁₂ ), so that i duplicated chromosomes are included in a single object is the number of objects)
[Equation 2]

(In Equation 2 above,
The D _F (Ab) is the number of abnormal chromosomes,
The D ₁ (Cr) is the number of chromosomes recognized through the first neural network,
The D ₂ (Ab) is the number of abnormal chromosomes recognized through the second neural network).
The method of claim 1,
The first neural network is
a first labeling unit for labeling the chromosome;
a first data enlargement unit for enlarging the first data of the chromosome; and
A chromosome reading system for diagnosing radiation exposure based on deep learning that includes; a first learning unit that educates the first data through machine learning.
The method of claim 1,
The second neural network is
a second labeling unit for labeling the abnormal and the duplicate chromosomes;
a second data enlargement unit for enlarging second data of the abnormal and duplicate chromosomes; and
A chromosome reading system for diagnosing radiation exposure based on deep learning that includes; a second learning unit that educates the second data through machine learning.
delete delete The method of claim 1,
The reading unit cannot read if the number of chromosomes does not satisfy Equation 3 below, and if the number of chromosomes satisfies Equation 3 and contains an abnormal chromosome, it is determined as abnormal, and if it does not, it is determined as normal. Chromosome reading system for diagnosis of learning-based radiation exposure:
[Equation 3]

(In Equation 3 above,
Wherein N is the number of chromosomes,
wherein I is 0 to 5).
A chromosome reading system for diagnosing radiation exposure based on deep learning,
Recognizing a chromosome through a first neural network;
recognizing abnormal and duplicate chromosomes through a second neural network;
analyzing the number of normal, abnormal and duplicate chromosomes based on the recognized information; and
Including; determining the chromosome as unreadable, normal or abnormal using the analyzed information;
The step of analyzing the number of chromosomes uses Equation 1 below,
The step of analyzing the number of abnormal and duplicate chromosomes is to use Equation 2 below, a reading method using a deep learning-based radiation exposure diagnosis chromosome reading system:
[Equation 1]

(In Equation 1 above,
The N _T is the number of chromosomes,
The N ₁ is the number of chromosomes recognized through the first neural network,
The N _i (S ₁₂ |C) is a chromosome in which the region recognized through the second neural network meets the criterion C and the region recognized by the first neural network (S ₁₂ ), so that i duplicated chromosomes are included in a single object is the number of objects)
[Equation 2]

(In Equation 2 above,
The D _F (Ab) is the number of abnormal chromosomes,
The D ₁ (Cr) is the number of chromosomes recognized through the first neural network,
The D ₂ (Ab) is the number of abnormal chromosomes recognized through the second neural network).
8. The method of claim 7,
Recognizing the chromosome
labeling the chromosome;
enlarging the first data of the chromosome; and
A reading method using a chromosome reading system for diagnosing radiation exposure based on deep learning, including; educating the first data through machine learning.
8. The method of claim 7,
Recognizing the abnormal and duplicate chromosomes
labeling the abnormal and the duplicate chromosomes;
enlarging second data of the abnormal and duplicate chromosomes; and
A reading method using a chromosome reading system for diagnosing radiation exposure based on deep learning, including; educating the second data through machine learning.
delete delete 8. The method of claim 7,
In the determining step, if the number of chromosomes does not satisfy Equation 3, it is impossible to read, and if the number of chromosomes satisfies Equation 3 and contains an abnormal chromosome, it is abnormal, and if it does not, it is determined as normal. , Reading method using deep learning-based chromosome reading system for radiation exposure diagnosis:
[Equation 3]

(In Equation 3 above,
Wherein N is the number of chromosomes,
wherein I is 0 to 5).

Images